The Interactive Advertising Bureau (IAB) published version 2 of its AI Transparency and Disclosure Framework on August 18, updating the January release to account for disclosure laws now in force. The document argues for risk-based labeling rather than a blanket rule, and it cites a number as evidence: an outside study found that telling consumers an ad was made with generative AI cut click-through rate by 31.5%.
What the framework recommends
The framework’s own position, quoted by MediaPost from the document: “labeling applied to AI involvement that poses no risk of deception penalizes advertisers without protecting consumers.” It sets three thresholds for whether a label belongs on a piece of creative: potential for deception, material impact, and clear consumer expectations. Overusing labels, it warns, creates “label fatigue” that reduces their effectiveness. Caroline Giegerich, VP, AI at IAB, framed the goal in trust terms: “Trust is everything between a brand and its customers, and being honest about AI is part of earning it.”
Which AI uses does the IAB framework say to label?
The framework recommends labels for AI-generated images and video built from a prompt, synthetic avatars and photorealistic AI influencers, digital twins of deceased people, digital twins of living people placed in fabricated scenarios, some synthetic voices, and AI chatbots that could be mistaken for a human. The exempt list covers: routine photo retouching, stylized or cartoon imagery, authorized voice clones of real people for scripted commercial content, generic voiceover, background music, and ad copy, including headlines, product descriptions and translations.
| Recommended: label it | Exempt: no label |
|---|---|
| AI-generated image or video from a prompt | Color correction, lighting, dust removal |
| Synthetic avatar or photorealistic AI influencer | Cartoon or stylized imagery, mascots |
| Digital twin of a deceased person | Authorized voice clone, scripted ad |
| Living person’s digital twin in a fabricated scenario | Generic voiceover, identity immaterial |
| Some synthetic voices | Background music, noise reduction |
| AI chatbot that could pass for human | Ad copy, headlines, translations |
The study behind the 31.5% figure
The finding is not the IAB’s own research. It comes from a working paper by NYU Stern and Emory University researchers, Ghose, Lee, Todri and Adamopoulos, titled “The Impact of Visual Generative AI on Advertising Effectiveness.” NYU Stern published a research highlight on it in November 2025, about nine months before the framework cited it; peer-review status is not established. The researchers compared human-made ads, AI-modified human ads and fully AI-generated ads in a lab experiment and a field study on the Google Display Network, per PPC Land’s reporting. Fully AI-generated ads raised click-through by up to 19% against human-made controls; AI-modified ads showed no meaningful gain. The study reports an “AI-generated” or “AI-edited” label cutting click-through by about 1.17 percentage points against unlabeled human ads, the 31.5% relative drop. The sources give no sample size or study duration, and none claims the effect replicates beyond the conditions tested.
The label itself
Where the framework does call for a label, it specifies the format. Advertisers can use a monochrome sparkle glyph at Unicode codepoint U+2728 or the plain text “AI-generated,” at the WCAG AA contrast ratio of 4.5:1. Video carries the label in the first frame, persisting throughout; audio states it before or immediately after the AI segment, repeated at least once if the ad runs over 60 seconds; images carry it nearby where feasible. Metadata carries C2PA provenance credentials with custom IAB assertions. Marketing Dive notes US advertisers can use the standardized sparkle icon or a text label, while an EU icon is still unfinalized. The rollout, per PPC Land’s account of the document, is a recommended sequence, not a deadline: name an AI Disclosure Lead within 60 days, pilot the labels, then scale to automated flagging over roughly two years.
Where this sits against actual law
The framework responds to a run of recent disclosure laws in the EU, California, New York, South Korea, China, India and Vietnam. Graham Wilkinson, EVP, Chief Innovation Officer at Acxiom, put it plainly: “Regulators in New York, California, South Korea and the EU are no longer asking whether AI disclosure matters.” None of that makes the IAB’s document binding; it is a trade-body recommendation, not a law. On the EU side, elsop’s own reporting on Article 50 found the duty reaches four narrow scenarios split between providers and deployers, not a blanket requirement to label AI-generated content, so routine AI-drafted ad copy generally sits outside the duty, provided it avoids health, consumer safety or sustainability claims. The framework’s exempt list for text and copy lines up with that reading.
One platform already drew a version of this line on its own. Snapchat’s July decision on AI-generated video keeps wholly AI-generated video publishable on Spotlight but drops its recommendation eligibility, whether the clip came from Snap’s own tools or a third party’s; video enhanced or edited with Snapchat’s own AI creative tools keeps that eligibility and carries a transparency indicator that is Snapchat’s own platform choice, not a response to the EU rule.
None of this is a deadline for anyone reading it this week. It is a sorting rule. Under it, the AI-retouched product shot going out Friday carries no sparkle; the synthetic influencer fronting next quarter’s campaign does.
